Source-linked AI summary
An Overview of Melanoma Detection in Dermoscopy Images Using Image Processing and Machine Learning
Nabin K. Mishra, M. Emre Celebi
TL;DR
Melanoma is important to detect early because it is a rising, potentially curable skin cancer, while current diagnosis can involve benign mimics and unnecessary biopsies. The paper reviews computerized dermoscopy analysis through lesion segmentation, clinical feature segmentation, and machine-learning classification, concluding that the approach has potential for earlier, cost-effective detection but requires further progress in accuracy and reproducibility.
Problem
Melanoma is a rising and potentially curable skin cancer, creating a need to distinguish it from benign mimics using dermoscopy images.
Method
The paper reviews lesion segmentation, clinical feature segmentation, and classification of attributes generated from segmented dermoscopy features.
Results
Computerized dermoscopy analysis has potential to support earlier melanoma detection and reduce unnecessary biopsies, but widespread implementation awaits improved accuracy and reproducibility.
Takeaways & Limitations
Dermoscopy-based image processing and machine learning could support fast, accurate, cost-effective melanoma detection in clinics or at home.
Takeaways & Limitations
Lesion segmentation remains constrained by poor contrast, artifacts, illumination variation, and broad variation in lesion appearance.
Abstract
from arXiv · showhide
The incidence of malignant melanoma continues to increase worldwide. This cancer can strike at any age; it is one of the leading causes of loss of life in young persons. Since this cancer is visible on the skin, it is potentially detectable at a very early stage when it is curable. New developments have converged to make fully automatic early melanoma detection a real possibility. First, the advent of dermoscopy has enabled a dramatic boost in clinical diagnostic ability to the point that melanoma can be detected in the clinic at the very earliest stages. The global adoption of this technology has allowed accumulation of large collections of dermoscopy images of melanomas and benign lesions validated by histopathology. The development of advanced technologies in the areas of image processing and machine learning have given us the ability to allow distinction of malignant melanoma from the many benign mimics that require no biopsy. These new technologies should allow not only earlier detection of melanoma, but also reduction of the large number of needless and costly biopsy procedures. Although some of the new systems reported for these technologies have shown promise in preliminary trials, widespread implementation must await further technical progress in accuracy and reproducibility. In this paper, we provide an overview of computerized detection of melanoma in dermoscopy images. First, we discuss the various aspects of lesion segmentation. Then, we provide a brief overview of clinical feature segmentation. Finally, we discuss the classification stage where machine learning algorithms are applied to the attributes generated from the segmented features to predict the existence of melanoma.
Introduction
Melanoma is a potentially curable skin cancer whose early detection is important because incidence and mortality remain substantial. Dermoscopy and public image databases support computerized analysis through lesion segmentation, feature segmentation, and classification.
- Melanoma usually begins in skin melanocytes and can vary in pigmentation, elevation, texture, and clinical structures.
- Early melanoma detection matters because melanoma incidence is rising and early-stage disease is more readily curable.
- Dermoscopy acquires magnified, illuminated skin images that improve visualization of skin spots and lesions.
- Public dermoscopy databases enable development and comparison of algorithms organized around lesion segmentation, feature segmentation, and classification.
Lesion Segmentation
Lesion segmentation separates the lesion from surrounding skin so global and local features can be analyzed. It is difficult because images contain weak contrast, artifacts, lighting variation, and substantial lesion diversity, requiring preprocessing, segmentation, and post-processing strategies.
- Lesion Segmentation: Lesion segmentation separates the region of interest from normal skin and supports morphological analysis, local feature segmentation, and border-based measurements.
- Lesion Segmentation: Segmentation is challenged by poor contrast, skin-tone variation, artifacts, non-uniform illumination, vignetting, and variation in lesion color, texture, shape, size, and location.
- Lesion Segmentation: Pre-processing may correct lighting and color variation, enhance contrast, smooth images, remove hairs, reduce vignetting, and localize lesions.
- Lesion Segmentation: Segmentation approaches include thresholding, clustering, region identification, active contours, edge detection, fuzzy logic, supervised learning, graph theory, and probabilistic modeling.
- Lesion Segmentation: Post-processing can merge regions, smooth boundaries, remove peninsulas and islands, and expand borders.
- Lesion Segmentation: Evaluation commonly compares automated borders with standards formed from multiple human expert segmentations, using subjective or objective criteria.
Feature Segmentation
Clinical feature segmentation identifies global, local, or distributed structures within lesions for diagnostic analysis. Its processing resembles lesion segmentation but must accommodate feature-specific appearance, artifacts, and multiple output segments.
- Feature Segmentation: Clinical features may be global, local, or distributed across multiple lesion locations, so feature segmentation often produces multiple segments.
- Feature Segmentation: Feature segmentation uses color, texture, shape, structure, relative size, location, and feature distribution as diagnostic attributes.
- Feature Segmentation: Pre-processing is feature-dependent and may involve color correction, lighting correction, enhancement, sharpening, blurring, or transformations.
- Feature Segmentation: Hair and gel masks can exclude occluded regions during pre-processing or remove spurious feature segments during post-processing.
- Feature Segmentation: Lesion-segmentation algorithms can also segment clinical features, but channel selection and post-processing must match the target feature.
- Feature Segmentation: Evaluation may include binary success or failure when the target feature is absent from an image.
Feature Generation and Classification
Melanoma prediction is a binary classification task using attributes generated from lesion and clinical feature segmentation. These attributes are supplied to classifiers and evaluated by accuracy, sensitivity, and specificity, with particular emphasis on detecting melanoma while identifying benign lesions.
- Feature Generation and Classification: Lesion and feature segmentation provide the preliminary steps for generating attributes used in benign-versus-malignant classification.
- Feature Generation and Classification: Global attributes include lesion diameter, symmetry, irregularity, eccentricity, color, and texture derived from borders or lesion areas.
- Feature Generation and Classification: Clinical attributes can include color, texture, shape, size, location, and surrounding-region information from segmented features.
- Feature Generation and Classification: Classifiers such as artificial neural networks, support vector machines, logistic regression, decision trees, and ensembles can distinguish melanoma from benign lesions.
- Feature Generation and Classification: Evaluation uses accuracy, sensitivity, and specificity, prioritizing high sensitivity while improving specificity and overall accuracy.
Conclusion
Computerized melanoma detection from dermoscopy images depends on a pipeline of lesion segmentation, feature segmentation, feature generation, and classification. Properly addressing image artifacts and segmentation quality is presented as essential for accurate, potentially accessible diagnosis.
- The workflow comprises lesion segmentation, feature segmentation, feature generation, and classification.
- Correcting dermoscopy-image aberrations and artifacts is crucial for achieving a correct diagnosis.
- Acceptable lesion-segmentation tolerance supports precision in feature segmentation and helps maximize classification accuracy.
- Fast, accurate, cost-effective computer analysis could support melanoma detection in clinics or even at home.